Less Is More . . .

AMD Announces X2 Processor With Only 256K Cache Per Core

Last year, AMD had dual-core Semprons on its roadmap. They disappeared rather quickly, or did they?

For practical purposes, this is a dual-core Sempron. This is not a bad thing, performance will probably be 3-5% lwss than a 3800+, but that’s still good enough to be competitive with a 3.2-3.4GHz Presler.

Why are they doing this?

Keep in mind that as long as AMD is making 90nm processors, every dual-core processor made reduces the total number of CPUs AMD can make, simply because dual-cores are bigger. By reducing cache, AMD is trying to reduce that penalty.

Yes, they’re getting additional manufacturing capacity from Fab 36 and Chartered, that’s the other way they’ll cover the rest of the size penalty.

Cost probably plays a bit of a role, too. The initial cost is supposed to be about $140, but that’s not going to hold. By the end of the year, the street price of a base level “regular” dual-core CPU will probably be around $100.

No doubt this is meant for the OEM market, but if I were AMD, I’d do whatever I could to make sure there were plenty of good, cheap AM2 mobos around, forget the SLI stuff. A big difference between AMD and Intel in the price of the mobo could be a decisive factor for a lot of people (including those Sixpacks in your life who need an upgrade).

Intel Comes Up With “Core Multiplexing”

“Core Multiplexing Technology” is Intelese for “when an app is single threaded, disable one core and let the other one use all the cache.”

While we suspect AMD’s reversing hyperthreading will be no savior, we know “core multiplexing technology” isn’t going to do much for any Intel chips. Double the cache on a single core, and you’ll get single digit overall performance improvement, and using another chip’s cache can hardly be better and likely will be worse.

Unless Intel somehow also overclocks the remaining core while doing this (and wouldn’t that be ironic), this seems little more than a gimmick to match AMD’s.


Be the first to comment

Leave a Reply